MLP-Mixer: An all-MLP Architecture for Vision - 42Papers